Experimental Design Details
We analyze a duopolistic price-inventory newsvendor game in which two retailers first set prices, then make inventory decisions after
observing each other’s chosen prices. In our model, demand function each retailer faces consists of a deterministic component, allocated based on relative prices, and a stochastic component reflecting market uncertainty. In these frameworks, a segment of consumers actively search for the lowest price and always purchase from the retailer offering the lower price. The remaining “non-searchers” simply buy from the first store they encounter, provided the price is below their reservation value. Consequently, the lower-priced retailer systematically secures a larger share of market demand, regardless of the absolute magnitude of the price difference, while the higher-priced competitor serves a smaller, less price-sensitive segment.
We choose experimental parameters according to Proposition 1 to generate different equilibrium predictions for pricing and inventory behavior across treatments. In all treatments, the reserve price r is fixed at 12. The base demand levels are set at dH = 100 for the high-demand market segment and dL = 50 for the low-demand segment. After the price stage, each newsvendor learns whether he or she has secured the high- or the low-demand segment but does not yet observe the realized demand. Demand realizations in each period follow a uniform distribution, represented as di ∼ U(˜ d − x, ˜ d + x). This means, in the LU treatments (x = 20), higher-priced participants face a demand interval of [30, 70] and lower-priced participants face the high-demand interval of [80, 120]. In the HU treatments (x = 40), these intervals widen to [10, 90] and [60, 140] for higher-priced and lower-priced participants, respectively.
The experiment was programmed and conducted in oTree (Chen et al., 2016). In total, we ran eight sessions, with two sessions for each of the four treatments. A total of 192 subjects, with 24 per session, participated in the experiment. Participants were undergraduate and graduate students from Harbin Institute of Technology in Harbin, China. Each subject provided written consent and participated in only one session, and made newsvendor decisions for 50 rounds.
At the beginning of each session, the experimenters distributed the printed instructions and read them aloud. The instructions included numerical examples and practice questions to ensure that subjects understood how token earnings were calculated. After confirming comprehension, the experiment proceeded on computers.
Within each session subjects were randomly assigned to fixed groups of four. These groups remained intact for the entire experiment and serve as independent observations. In every round two members of a group were randomly matched to form a duopoly, and identities were not revealed.
Each round had two stages. In stage 1 both sellers chose a price. The admissible price grid had one-decimal-place increments: 3.0 to 12.0 tokens in the HM treatments and 9.0 to 12.0 tokens in the LM treatments. After prices were posted, each seller learned whether he or she had won the high-demand segment. In stage 2 the sellers chose inventory levels. Order quantities were integers from 0 to 120 tokens in LM HU and from 0 to 140 tokens in LM LU. Unsold stock was discarded at the end of the round.
When both decisions were complete, the program drew demand, calculated profits, and displayed feedback, including their selected selling price, inventory quantity, realized demand, round profit, and the accumulated earnings. At the end of the experiment, participants completed a brief survey on demographic information, such as gender, major, school year, and prior experience with laboratory decision-making experiments.